Prostate cancer, the most commonly diagnosed cancer in U.S. males, displays a broad spectrum of clinical behavior from indolent to fatal. Our current tools for prognostication are crude at best, leaving patients and their doctors struggling to decide which tumors need treatment, and, if so, how aggressively they should be treated. New molecular genetic markers of prostate cancer prognosis, ideally derived from an improved understanding of the molecular underpinnings of prostate cancer, are desperately needed. We have identified 3 distinct subtypes of prostate cancer by comprehensive gene expression profiling that appear to be clinically relevant and that would not be predicted by traditional clinical and pathological parameters. From these expression profiles, we have identified genes correlated with these subtypes that provide independent prognostic information on patients after surgery. We propose to validate and expand this gene expression dataset to identify additional markers of prognosis. We will analyze tumor samples from patients who have recurred after surgery and compare their gene expression patterns to matched samples from patients who have not recurred. Prognostic markers will be identified from intrinsic features of the data (tumor subclasses) and from supervised and semi-supervised analyses of data from recurrent and nonrecurrent patients. We will also perform genome-wide analysis of DNA copy-number changes using microarray-based comparative genomic hybridization to evaluate the extent to which DNA changes contribute to or underlie the molecular subtypes of prostate cancer. Finally, the clinical significance of individual transcript alterations, their cognate proteins, or regions of DNA amplification or loss will be evaluated on tissue microarrays with associated detailed clinical data and long-term follow-up. The proposed work will help interpret our prostate cancer gene expression dataset and translate findings from this work into clinical practice.